> an increase in thin papers of narrow scope, as well as ‘salami’ papers, where a single work is broken up and submitted as a set of smaller papers. There is also a marked increase in dense, AI-written papers. AI tools are making it easy for authors to flood arXiv [and others with] low-value papers.
![Identity: What Characterizes Connectionism?
The model … represents an essentially empiricist approach to
perception [with] an optical input, and a printer or set of signal lights
as an output. [A]fter a period of training, the system will exhibit
capabilities for discrimination, association, and stimulus generalization.
[T]his is the first time that a set of theoretical principles will have been
clearly proven to generate a perceptual capability, in a system of
completely known structure. (Rosenblatt, 1959, pp. 296–297)
C-connectionism bases its identity in part on showing that connec-
tionist models can account for phenomena that do, or did, not appear
to be easily composable into computations carried out by smaller
units, such as the so-called neurons and their connection weights
in ANNs (see Table 1, especially rows Goal, Question, and Training).
C “has demonstrated that a great deal of information is latent in the
environment and can be extracted using simple but powerful learning
rules” (Elman et al., 1996, pp. xii–xiii). In many ways, this is a
theoretical point about the nature of what the modeled organism is
doing, that is humans are able to learn statistical regularities from the
environment, which C-connectionism rightly takes to mean that the
resulting learning is by virtue of the training set and of the learning
algorithm (Guest et al., 2020). That is, they set their scientific sights on
understanding if their connectionism can, given the ANNs they build
and the training sets they train their model on, give rise to what they
see people do. C-connectionists, like Elman et al. (1996) and many
others, explicitly reacted to claims of innateness (however construed)
of capacities, asserting and showing (according to their standards of
evidence) that appealing to innate properties of neurocognitive sys-
tems is not required.1
On the other hand, what characterizes M-connectionism is that its
identity is based on (a) deep ANN models, which are framed as
human-lik…](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreiefimaldrgopztrw5gquhalz6fm74fazl7murpdtdqe74izwdfs5y)
![Finally, (d) M-connectionism implicates brain areas, and neu-
roscience generally, more often than C-connectionism and with an
agenda entangled with so-called neuro-/bio-plausibility (see row
brain, in Table 1). Let us contrast again with a typical C stance on
this: “Neural plausibility should not be the primary focus for a
consideration of connectionism” (M. S. C. Thomas & McClelland,
2008, pp. 28–29, also Smolensky, 1988, but cf. Elman et al., 1996;
McLaughlin & Warfield, 1994; Stinson, 2020). Additionally,
appealing to innateness and nativism—that is that “aspect[s] of
cognition [… ] must be innate, or, (at the very least) subject to
powerful biological constraints” (Elman et al., 1996, p. 240)—is not
ruled out by M-connectionism. In fact, such constraints are ap-
pealed to, or seen as imperative to include in models, using phrases
such as “inductive bias” (a concept which has been around in
machine learning for a while, including with respect to ANNs;
Gordon & Desjardins, 1995; Pavlick, 2023). Inductive biases like
innate capacities are “not learned” (Goldberg, 2008; see footnote 1).
And as such stand in stark contrast to requests from C practitioners
to not “pre-wire structure into [the] mechanism if it can [be obtained]
for free from the environment” (Plunkett, 2001, p. 193), that is the
training set.
In both types of connectionism, although especially in M,
there emerge similar entanglements between observations and the
so-called “bridging” of levels, which proposes that, for example,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
All rights, including for text and data mining, AI training, and similar technologies, are reserved.
1 As Elman et al. (1996) and others acknowledge, innateness and nativism
are not clear-cut concepts and the polarized, or even wrong framings of the
nature/nurture…](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreif2heeccnwtisvo66ghvzn5fbyfxrg4noqfsrx7hf2e43q7hwla3q)
![thus mainstreaming the use of mental representation (e.g. Favela
and Machery, 2023; cf. Amundson, 1983). Notwithstanding,
and recalling the careful threading of the needle described in
Box 1:
It is a paradox that the “Tolmaniacs” [Tolman’s stu-
dents] from Berkeley who tend to speak of rats as
“little furry people” are much more likely to search
for central neural mechanisms than S-R[i.e. stimulus-
response] theorists who speak of rats as “little furry
machines.”
John Garcia (1976, p. 81)
To recapitulate our main point, herein we have built the case
for cognitive map needing careful use, mindful deployment
when verbally and formally theorising. And so in the case of
neurocognitive mental representations cognitive maps can be
“systematically misleading” per Ryle, i.e.
the sense in which such quasi-ontological statements
are misleading is not that they are false and not even
that any word in them is equivocal or vague, but only
that they are formally improper to the facts of the
logical form which they are employed to record and
proper to facts of quite another logical form](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreih3uk5rjryjp32uttxxvvwek4q45kssh6lc7xzdmh6hslen6nrtx4)
![These problematic beliefs exist on a spectrum, from the assumption that AI products can help in mental health (e.g. Dehbozorgi et al., 2025; Siddals et al., 2024; Wellcome, 2025) to the
assertion that a chatbot can be a therapist (Kilgore, 2025), which
is a regulated profession (e.g. European Federation of Psychologists’ Associations, 2025) with codes of ethics (e.g. American Psychological Association, 2017; British Psychological Society, 2021).
Proponents of these beliefs use the guise of labour shortages and
the global mental health crisis (Kaplan, 2024), as cover to deskill
(for an alarming case, see Budzyń et al., 2025; also Akingbola et
al., 2024; Lebovitz et al., 2021), contributing to the polycrises the
technology industry and their allies uniquely profit from (McConnell & Jacobs, 2025).
All AI, and indeed any technology in mental or other healthcare settings requires “keeping users safe [and this in turn] requires substantial input from clinicians and careful planning to
reduce risk.” (Abrams, 2025) This is key because it is not possible to keep people safe without a highly qualified human-in-theloop (Amironesei et al., 2021; El-Mhamdi et al., 2022; T. Liu et
al., 2025; Salecha et al., 2024; Sharma et al., 2025; Weidinger et al.,
2022) and even then the risk of deskilling is present (e.g. Budzyń
et al., 2025; recall section 4: Outsourcing Programming to Companies). Relatedly, and taking the example of the therapist-client
relationship, dehumanising parallels encoded in expressions such
as: “the country needs all the quality therapists we can get —
be they human or bot” (Riddle, 2025) cannot become normal
in our scientific discussions. No therapist under proper ethical
functioning would cause their client to be addicted to their therapy (Huntington, 2025) nor would they, as in the Replika example (The Luddite, 2024), introduce a sexual relationship between
them and their client (Pettit, 2024b). Any such suggestions to
use a chatbot as a therapist would go aga…](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreifixyivu7iw2adfj63dmenyvvzpslcgl2h7sl6ngdkzvvz23qkeri)
![Technology is giving life
the potential to flourish
like never before...
[half-dead tree image]
...or to self-destruct.
Let's make a difference!](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreiaul4ts7stnot7mmhk66mp7petxl4nimzovfu6dqlo242ewuqchz4)
![There it is a definite social relation between [people], that assumes, in their eyes, the fantastic form of a relation between things. In order, therefore, to find an analogy, we must have recourse to the mist-enveloped regions of the religious world. In that world the productions of the human brain appear as independent beings endowed with life, and entering into relation both with one another and the human race. So it is in the world of commodities with the products of [people]’s hands.
(Marx, 1867)
The reasoning problems become evermore severe in our misunderstandings of these most modern machines (Guest & Martin, 2023). As Taina Bucher (2018, p. 50) explains: “When a machine runs smoothly, nobody pays much attention, and the actors and work required to make it run smoothly disappear from view (Latour, 1999).” A next step in this devolution and devaluation of cognitive labour is that now the user too, deskilled and displaced, also disappears from view. What voice does the human, now reduced to only a user, have if their verbal (column 1, Figure 4) and visual (column 2) expressions are just the copy-pasted output of a device that performs patchwor](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreiglgkond7er3fuisrkbcllwojnwobpcevsmqyuejn2e7ur7s7mnru)
![metatheoretical
calculus
Metatheoretical calculus captures the reasoning we as practitioners of our science perform over data,
models, and theories. This comprises, but is not limited to: how we relate theories to each other, our
adjudications within and between different theoretical accounts, how we infer theoretical or modelling
successes, and what we deem virtuous or vicious (Guest, 2024; Guest & Martin, 2023). Labelling
these ways of thinking promotes critique of, and improvements to how psychological, neuro-, and
cognitive science is carried out (e.g., Guest & Martin, 2026; Guest, Scharfenberg & van Rooij, 2025).
artificial
intelligence (AI)
Herein we use AI to mean the industry, its concomitant logics and technologies, and the related
research programmes that go under this heading. AI "impedes our theorising about phenomena and
systems under study [because] we are interested in human-understandable theory and theory-based
models, not statistical models which provide only a representation of the data. Scientific theories and
models are only useful if [we understand them and] they connect transparently to research questions."
(Guest & van Rooij, 2025, Table 1)
For a more exhaustive list of meanings of AI see: Table 1 in van Rooij et al. (2024) ; for separating
such technologies as a function of harms see: Guest (2026b); and for a general overview of how to
think about psychology and AI see: Guest and van Rooij (2025) and van Rooij and Guest (2026).
For our analyses overall see: Erscoi et al. (2023), Forbes and Guest (2025), Guest (2026a), Guest
and Martin (2023, 2025, 2026), Guest, Nuñez Hernández and Blokpoel (2026), Guest, Suarez and
van Rooij (2025), Guest, Suarez et al. (2026), Ray and Guest (2026) and Spanton and Guest (2022)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreiawuhxigjeyvtewmlzskrcm46rynw4bm7vjqhtnnrqufoadc6ecuq)
!["[A]ll science would be superfluous if the outward appearance and the essence of things directly coincided. (Marx, 1894, p. 592)"
This is the logic of AI, of fascism, of sexism, of racism. That is, because things look a certain way, that therefore they are a certain way. Internal properties are wrongly derived from their superficial appearance. "I see that somebody has long hair, therefore I can conclude they are a woman." Or worse still: "I see the external features consistent with (what I think is) woman, and I conclude she must be stupid" — a fractal wrongness that starts with assuming that what you (naively, unquestioningly, think you) see is what you get when it comes to complex organisms. Further down in the same paper, we warn:
"Just because a model correlates with neural and behavioral data, it is not sufficient for us to infer that the model is performing cognition: correlation does not imply cognition."
Again, the core logics of AI are the same as racism and sexism, which is why these issues pop up again and again in these systems: it involves taking the simple shallow features of things we see, like skin colour, or hair length, or whatever other superficial property, and moving to very flawed conclusions. There is something addictive to this form of sheer intellectual laziness — as well as structurally beneficial to those select few within capitalism, white supremacy, and patriarchy — to this simplistic logic. It's easy](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreiaim77cybhtnrdwgc4ts5igcbogmdiwblb32tjaq65xu7s5bpmmoe)